SelvarMix-package: Regularization for variable selection in model-based...

In masedki/SelvarMix: Regularization for variable selection in model-based clustering and discriminant analysis.

Description

SelvarMix is a package where a regularization approach of variable selection is considered in model-based clustering and discriminant analysis frameworks. First, this procedure consists of ranking the variables with a lasso-like procedure. Second, the method of Maugis et al (2009, 2011) is adapted to define the role of variables in the two frameworks. SelvarMix provides a faster variable selection algorithm than the backward stepwise or forward stepwise algorithms of Maugis et al (2009), allowing us to study high-dimensional datasets.

Details

Package:	SelvarMix
Type:	Package
Version:	1.0
Date:	2014-04-03
License:	GPL-3 + file LICENSE
LazyLoad:	yes

The general purpose of the package is to perform variable selection in model-based clustering and discriminant analysis. It focus on model-based clustering, where the clusters are assumed to arise from Gaussian distributions. The most achieved model in model-based clustering has been proposed by Maugis et al (2009). This so-called SRUW modeling considers three roles of variables: one variable my belong to the relevant clustering set S, the redundant variable set U or the independent variable set W. Moreover, the redundant variables may be explained by a subset R of the relevant variables S. In order to avoid the slow of this algorithm when data with numerous variables are studied, the SelvarMix procedure is proposed. It proceeds in two steps: First, the variables are ranked using a lasso-like procedure analogous to the one of Zhou et al (2009); second, the SRUW procedure is run on this ranked set of variables.

Author(s)

Author: Mohammed Sedki, Gilles Celeux and Cathy Maugis-Rabusseau

References

Maugis, C., Celeux, G., and Martin-Magniette, M. L., 2009. "Variable selection in model-based clustering: A general variable role modeling". Computational Statistics and Data Analysis, vol. 53/11, pp. 3872-3882.

Maugis, C., Celeux, G., and Martin-Magniette, M. L., 2011. "Variable selection in model-based discriminant analysis". Journal of Multivariate Analysis, vol. 102, pp. 1374-1387.

Zhou, H., Pan, W., and Shen, X., 2009. "Penalized model-based clustering with unconstrained covariance matrices". Electronic Journal of Statistics, vol. 3, pp.1473-1496.

Sedki, M., Celeux, G., Maugis-Rabusseau, C., 2014. "SelvarMix: A R package for variable selection in model-based clustering and discriminant analysis with a regularization approach". Inria Research Report available at http://hal.inria.fr/hal-01053784

Examples

## Not run: 
## Simulated data  example as shown in Maugis et al. (2009) (correlated scenario 2) 
## n = 2000 observations, p = 14 variables 
require(Rmixmod)
require(glasso)
data(scenarioCor)
data.cor <- scenarioCor[,1:14]
labels.cor <-scenarioCor[,15]


lambda <- seq(20,  100, by = 10)
rho <- seq(1, 2, length=2)
hybrid.size <- 3
models <- mixmodGaussianModel(family = "spherical", equal.proportions = TRUE)
regModel <- c("LI","LB","LC")
indepModel <- c("LI","LB")

## variable selection in model-based clustering  
nbCluster <-  c(3,4)
criterion <- "BIC"
simulate.cl  <- SelvarClustLasso(data.cor, nbCluster, lambda, rho, hybrid.size, 
                                 criterion, models, regModel, indepModel)



## variables selection in discriminant analysis
## training sample : n = 1900, p = 14 variables 
data.learn <- scenarioCor[1:1900,1:14]
labels.learn <-scenarioCor[1:1900,15]

## testing sample : n = 100, p = 14 variables
data.test <- scenarioCor[1901:2000,1:14]
labels.test <-scenarioCor[1901:2000,15]

lambda <- seq(20,  50, length = 10)

simulate.da <- SelvarLearnLasso(data.learn, labels.learn, lambda, rho, hybrid.size, 
                                models, regModel, indepModel, data.test, labels.test)

## End(Not run)

masedki/SelvarMix documentation built on May 21, 2019, 12:42 p.m.